Social visibility and stated confidence

Social visibility and stated confidence

Business Economics Organisation Economics

Master Thesis 10084932

Abstract

This thesis focuses on the stated confidence of individuals. People’s self-awareness (or overestimation) of abilities and skills influences individual decision-making. Socially visible estimation differs from private estimation. In case of socially visibility, people’s stated confidence includes a trade-off between overestimation – or accurate estimation of one’s actual performance. Overestimation can be a result of protection of social image or persistent overconfidence, whereas accurate estimation of abilities can be a result of lying-aversion or humility. An experiment with 103 primary school pupils tested the effect of social visibility on stated confidence, also in the relation with gender and age. This led to three conclusions: first, girls are more sensitive to the effect of social visibility on stated confidence compared to boys. Second, stated confidence of boys is only significantly higher compared to girls in private stated confidence. Finally, stated confidence differs significantly with age in private stated confidence.

Statement of Originality

This document is written by Student Lotte van de Lustgraaf who declares to take full responsibility for the contents of this document.

I declare that the text and the work presented in this document is original and that no sources other than those mentioned in the text and its references have been used in creating it.

The Faculty of Economics and Business is responsible solely for the supervision of completion of the work, not for the contents.

Content

1 Introduction 4

2 Literature review 5

2.1 Social image 6

2.1.1 Effect of monitoring behaviour 6

2.1.2 Trade-off between social image and lying-aversion 7

2.2 Overconfidence 8

2.2.1 Motives behind overconfidence 9

2.2.2 Social image and overconfidence 10

2.2.3 Effect of gender and age 11

2.3 Existing studies 12

3 Methodology 15

3.1 Experimental design and procedures 15

3.1.1 Background information of the data 15

3.1.2 Procedures and experimental design 16

3.2 Hypotheses 19

4 Results 20

4.1 Check for a normal distribution 21

4.2 Descriptive data 23

4.3 Interpersonal versus intrapersonal 27

4.4 Overestimation of subject’s performance 31

4.5 Differences in performance 33

4.6 Limitations 35

5 Discussion and conclusion 37

6 References 39

1. Introduction

This thesis studies the effects of social visibility on stated confidence. Self-awareness, or overconfidence, of abilities and skills of individuals influences decision-making which could be relevant information for organizations. For instance, the results of Barber and Odean (2001) showed that the higher overconfidence level of men resulted in lower returns in financial stock markets in comparison to women. However, the self-estimation of individuals is in principle private information (Ewers and Zimmermann, 2015), but what happens to people’s behaviour if their personal self-estimation becomes visible to others? The question is, if whether or not this type of personal information goes public influences people’s responses on (stated) confidence (Ewers and Zimmermann, 2015). It could lead to different conducts if individuals would know, in advance, that their stated confidence will be

anonymous (not visible) - or public (visible to other people). Possibly, people could

overestimate their personal level of stated confidence to protect their social image in case of social visibility (Ewers and Zimmermann, 2015). Individuals are aware of their actual performance level, but they pretend to be better than they really are on purpose. In certain conditions people gain benefits out of a positive bias: stated confidence does not necessarily correspond with people’s actual performance. Examples are found in many places: in politics, in the labour market and the workplace; overrating actual abilities could increase the

probability of getting more voters, a new job or promotion.

There are studies tending to the conclusion that this overconfidence is no incidental phenomenon: people are overconfident about their actual abilities in principle (Barber and Odean, 2001). This is not harmless. People’s persistent overconfidence of actual performance creates an inaccurate self-image, to the extent that this phenomenon tends to irrational

decision-making. In this case, persistent overconfidence can be harmful to organizations. However, overestimating of actual performance is not the only reaction of people if their self-estimation becomes socially visible; people could tell the truth about their believes or show humility with regard to their abilities and skills.

Economic literature on this issue is scarce. I have only found one other related study by Ewers and Zimmermann (2015). This study was published in an economic paper for the first time during my thesis research 25 March. Ewers and Zimmermann (2015) studies the misreporting of self-estimation due to social image protection.

The paragraphs above show the importance of the study of stated confidence. Whether it is private for people themselves or socially visible, whether men differ from women.

research-theme: Does social visibility influence stated confidence and does stated confidence depend

on gender and/or age?

We will find three conclusions in this thesis. First, young girls are more sensitive to the effect of social visibility on stated confidence in comparison to young boys. Second, the stated confidence of young boys is only significantly higher in comparison to young girls in case of reporting stated confidence in private (socially invisible). Finally, even at a very young age the stated confidence differs significantly with age in case of reporting stated confidence in private; the stated confidence of somewhat older individuals is significantly higher in comparison to the stated confidence of somewhat younger individuals

This thesis is structured as follows. The next chapter (2) will further elaborate on the relevant subject by a related literature review. I will explore the underlying theory about social image and overconfidence of individuals. This literature review illustrates the social relevance of a clear view on overestimation of individual’s abilities in labour organizations. Chapter 3 goes into the detail of my design of the experiment. Chapter 4 summarizes the results of the experiment. The thesis will finish with a discussion and conclusions.

2. Literature Review

In this chapter I will review the literature and existing studies on social image and

overconfidence of individuals in relation to stated confidence. In short, people overestimate their personal abilities in both cases (social image and overconfidence): people’s stated confidence is inaccurate in comparison to their actual performance level. This chapter will show that people will find themselves in situations where they gain from a more favourable image, but they are not automatically inclined to exaggerate (Section 2.1). This differs from the notion of “overconfidence”, where abilities of individuals are systematically

overestimated by their own judgement (Section 2.2). Situations on the labour market could provide some examples. For instance, personal abilities of (future) employees are not known beforehand in situations like applications or promotions (Ewers and Zimmermann, 2015). The fact that other people will find out about the personal stated self-confidence could influence the way in which people display their self-esteem (e.g. in personal assessments).

Overestimation of individual’s abilities is the most feasible probability in case of social visibility of personal stated confidence.

2.1 Social image

Most studies on social image investigate how a more social desirable outcome could be reached (Ewers and Zimmermann, 2015). Bénabou and Tirole (2006) for example provide a theory where charitable giving is seen as a prosocial behaviour. Prosocial in this case means that people do not only care about themselves, but also about the well-being of others

(Bénabou and Tirole, 2006). The experiment of Jones and Linardi (2012) studied whether prosocial behaviour changes in case of increasing social visibility. They concluded that women change their behaviour into the average behaviour of other individuals with social visibility; women do not want to attract attention based on their image and actions (Jones and Linardi, 2012). What other people think about a person influences how that person functions in daily life, as well at his/her workplace or within other social networks. In general, people want to be liked by others (Falk and Zimmermann, 2011; Bénabou and Tirole, 2006).

2.1.1 Effect of monitoring behaviour

Individuals respond on signals of being watched, even if these signals are very subtle or unknown (Bateson et al., 2006; Haley and Fessler, 2005). An example for a subtle signal is a display of eyes in a room in the studies of Bateson et al., (2006) and Haley and Fessler (2005), which is a subtle signal of feeling watched without a real observer. Individuals adjust their behaviour according to signals of observation (being observed), especially when their actions or decisions become socially visible. The change in behaviour depends on the degree in which individuals appreciate their social image (Bateson et al., 2006; Bénabou and Tirole, 2006).

The study of Falk and Kosfeld (2006) and Bateson et al. (2006) show the type of

adjustment in behaviour of individuals due to signals of feeling observed. The study of Falk

and Kosfeld (2006) showed a negative effect on people’s behaviour; people do not like to be controlled in the workplace and respond negatively by decreasing their productivity as a reaction on principal’s distrust. The study of Bateson et al. (2006), on the other hand shows, a change towards positive behaviour due to subtle signals of being watched in a field

experiment of collecting money. Additionally, laboratory studies show that this effect of positive behaviour in collecting money, caused by feeling monitored, does not change in case of increasing individual’s anonymity (Hayley and Fessler, 2005). Individuals behave more selfish without the feeling of being monitored in an anonymous setting (Dana et al., 2007; Levitt and List, 2007). Whether people’s actions and decisions become socially visible or not is an essential aspect in the reaction of individual’s behaviour (Levitt and List, 2007).

2.1.2 Trade-off between social image and lying-aversion

Level of confidence and people’s self-image of individual performance, abilities and skills is highly related to their functioning in organizations. People like to work with individuals with a high confidence instead of individuals with a low self-esteem (Bénabou and Tirole, 2002) and employers prefer to work with employees where the self-estimation corresponds with actual performances (Falk and Zimmermann, 2011). The correct self-estimation increases the value of accurate communication between individuals (Gneezy, 2005). Highly confident employees with a correct balance between self-estimation and actual performance are of great value to organizations.

Receiving positive- or negative feedback about performances improves someone’s self-image of actual skills and abilities. Feedback about performance increases people’s learning process towards future performances. Individuals who like their positive self-image, whether this self-image is accurate or not, would not revise their individual estimations about their performance (Köszigi, 2006) and hold on to their positive feeling of self-image.

Neglecting feedback reduces the learning process (Köszigi, 2006) and results in a(n) (over)biased self-image. Besides, people could pretend to be better than they really are to protect their social image, create a better reputation and a (too) positive self-image (Bénabou and Tirole, 2002) even if the self-estimation is inaccurate (Ewers and Zimmermann, 2015). In this case individuals act on purpose and with full knowledge and attach high value to how other people see them. This is a deliberate act: these individuals are aware of their actual performance, abilities and skills. Bénabou and Tirole (2002) explain the following three reasons why individuals want to create a positive self-image. At first, people draw strength from positive feelings instead of a “breakdown” from a negative self-image. Secondly, a positive and confident self-image increases the probability that other people in the network also belief in one’s abilities and skills. Finally, a positive self-image encourages individuals to follow their dreams and go for success. These three reasons of Bénabou and Tirole (2002) lead to the conclusion that a positive self-image corresponds with feeling good (with themselves). These positive feelings increase even more when other individuals in the surrounding network confirm this image (Leary et al., 1995). Having a positive self-image, showing confidence in the surrounding network and building a good reputation is stronger when individuals will meet each other again in the future (Levitt and List, 2007). However, a positive self-image does not necessarily mean that it leads to better decisions (Köszegi, 2006; Barber and Odean, 2001). Besides, this decision-making and behaviour of individuals is not

only influenced by the current situation, but also by individual’s past actions and experiences (Levitt and List, 2007).

Until now the different statements above could lead to the conclusion that individuals always try to exaggerate their skills and abilities, always try to impress their surrounding network with a good reputation; that individuals in general opt for the most positive (overestimated) reputation. But there are more relevant factors involved. Exaggeration of abilities and skills contradicts with the tendency that individuals try to avoid lying (lying-aversion) in general (Gneezy, 2005). In comparison to the expectations out of standard economic theory, individuals lie less frequently, communication includes more information (Lai and Lim, 2012) and acts are more in individual’s moral conviction (Levitt and List, 2007). This could reveal, or lead to, truthful information during the communication (Lai and Lim, 2012). More exactly, one’s self-estimation tends to correspond with the actual

performance in cases where lying-aversion dominates. In that case; image building is not to pretend to be better than you are, but showing the real self-esteem to others, where individuals stay close to their moral values. The true image of abilities, skills and performances is what, for example, principals want to discover in employees in case of long-term contracts.

The balance between these two tendencies will vary between (categories of)

individuals and groups. Individuals will make an internal trade-off between the value of their social image and their desire not to tell their true estimation of confidence. For example low productive individuals will pretend on purpose to be better than they really are to create better (work) opportunities (Ewers and Zimmermann, 2015). The creation of a better reputation is less necessary for high productive individuals; they will show their true self-estimation more often (Ewers and Zimmermann, 2015).

Unfortunately, the actual knowledge about individual’s abilities is generally private information (Ewers and Zimmermann, 2015). The labour market could gain efficiency if this information about performance and abilities becomes socially visible in the case of allocation of job applicants, work tasks or promotions of employees (Ewers and Zimmermann, 2015). In general personal assessments try to investigate a (unexpected) gap.

2.2 Overconfidence

Decision-making in organizations is influenced by the self-image of the employee’s abilities. Results of studies, mostly psychological, indicate that people are overconfident about their abilities (Barber and Odean, 2001) resulting in a biased report of their true abilities (Möbius et

al., 2011). A persistent biased self-image decrease the probability of successful decision-making, in daily life and at work.

2.2.1 Motives behind overconfidence

There are different definitions of overconfidence used in the literature regarding this subject. Moore and Healy (2008) explain the following three main definitions of overconfidence where personal self-image and the actual truth do not correspond with each other.

 First, people overestimate their skills and abilities.

 Second, people estimate their individual ability as “above average”.

 Last, people believe too often that their belief is the actual (and only) truth.

Overconfidence can create serious damages, when employees are overestimating their abilities at their workplace. Overconfidence of employees does not always lead to a better performance within an organization or that organization itself. For instance Barber and Odean (2001) shows that highly overconfident financial male investors create lower stock market returns in comparison to female investors.

Individual’s overconfidence increases in case of difficult and complex tasks, where success in performance is hard to reach, the outcome is uncertain and/or there is no feedback about the performance (Barber and Odean, 2001). Overconfident individuals act on

(overbiased) self-esteem; they believe that their ability is better than the general ability of others (Barber and Odean, 2001). They decide or act in contrast to expectations of economic theories on rational making and neglect the possible risks of harmful decision-making (Köszigi, 2006). Expectations of overconfident individuals are biased and not rational (Barber and Odean, 2001). Systematically overestimation of personal abilities can become a part of the personality of individuals; overconfident individuals develop a certain attitude of (systematically) underestimating potential risks in decision-making. In this case, individuals act not on purpose and not with truthful information about real performances, skills and abilities, because they do not know themselves truly. Having too many overconfident people within an organization will create an environment of overconfidence. This can lead to

disproportionate decision-making based on incomplete information or a perverse risk-attitude. This line of thought confirms the theory of behavioural economics, individuals do not only base their decisions on rational theories, but their decisions are influenced by a variety of

factors in human behaviour, such as a person’s feelings, moods and attitudes (Shiv and Fedorikhin, 1999; Niederle and Vesterlund, 2007).

In Section 2.1.2 we saw a trade-off between social image and lying-aversion. The contradictory tendency of the individual’s risk-attitude in case of overconfidence of Barber and Odean (2001), explained in the last paragraph, is the literature of showing humility in relation to abilities; self-confidence decreases if tasks involved become more difficult (Moore and Cain, 2007; Moore and Healy, 2008). People mainly focus their confidence level on their own past performance and neglect the fact that other people will find the task difficult as well (Moore and Cain, 2007; Moore and Healy, 2008). The focus on a person’s individual

performance will in this case lead to underconfidence; people think they did worse in comparison to others on a specific task.

This could be, for instance, an application test. When most people find the same questions of the application test difficult, they still compare their overall performance with others, before they judge their personal result. Otherwise they draw wrong conclusions about their performance. Furthermore, people know their personal abilities and should not only focus their performance level based on the tough(est) questions. Another example is found in exams of students; if those students are told beforehand that the exam will be very difficult and important for future prospects, students feel a high pressure and are less confident in comparison to announcements of a simple exam that everyone could accomplish (Moore and Healy, 2008). In contrast, less confident individuals could try harder to increase performances and boost their self-image (Köszigi, 2006).

2.2.2 Social image and overconfidence

Social image and overconfidence are the main elements described in this thesis, where both elements indicate that people overestimate their personal abilities and skills. In my view social image and (persistent) overconfidence are two different independent concepts; they can both explain people’s overestimation of actual performance. Ewers and Zimmermann (2015) see overconfidence as a consequence of individuals decisions to protect their social image or reputation. In my view the essential difference between the two concepts of overestimation lies in the awareness of overestimation of personal abilities. Table 1 summarizes the definitions of- and differences between the social image and overconfidence used in this thesis. The personal overestimation is “in control” by the specific individual in case of social image. In contrast, an overconfident individual does not control actions and decision-making

due to overestimation; this leads to persistent biased results of an individual’s stated confidence.

Table 1: Characteristics of social image and overconfidence.

Social image Overconfidence

- On purpose with full knowledge about true individual performance, abilities and skills

- Not on purpose. without full knowledge of individual performance, abilities and skills - can be (become) part of personality

- People’s behaviour differs in case of increasing social visibility

- Systematically behaviour; Always overconfident

- Positive feeling due to positive self-image - Attitude (arrogance) not based on rational decision-making

The separation of the two definitions of social image and overconfidence can be explained with an example of car drivers. There are car drivers who always exceed the speed limit, no matter in which situation. These drivers could be compared to persistent

overconfident individuals. There are also car drivers who vary their speed according to different situations. These individuals sometimes exceed the speed limit on purpose in a specific situation; these type of individuals are aware of the fact that they exceed the speed limit. These drivers can be compared to the state of social image.

We see that not every individual will make a trade-off between a positive social-image/reputation and on the other hand their desire to tell truth about their estimation of confidence level. This is only possible in cases where individuals are fully aware about their actual level of abilities and skills. In contrast, individuals that are not aware of their actual level of performance will not be able to make a choice when it comes to this kind of a trade-off.

2.2.3 Effect of gender and age

Gender is a factor in measuring people’s level of self-awareness (Barber and Odean, 2001). We see a lot of stereotype statements in this Section 2.2.3. I will describe them in briefly, without further elaboration.

In general, men have a higher level of overestimation of achievements compared to women (Niederle and Vesterlund, 2007; Barber and Odean, 2001; Möbius et al., 2011); women are less confident about their abilities than men (Niederle and Vesterlund, 2007).

However, when all performances are divided into different type of tasks it could be possible that women are more confident about a specific task compared to men; overestimation is task specific (Niederle and Vesterlund, 2007; Barber and Odean, 2001).

Overconfident individuals could improve their decision-making with feedback of their performance which was already mentioned in Section 2.1.2. However, many individuals protect themselves by ignoring (negative) feedback to maintain their positive self-image (Köszigi, 2006; Möbius et al., 2011). In the case of unavailable feedback, women are less confident about their performance than men (Barber and Odean, 2001). If individuals actually have to do deal with feedback, women are more touched by negative feedback in comparison to men (Niederle and Vesterlund, 2007).

The fact that there are less women in professional top positions can also be explained by overconfidence (Niederle and Vesterlund, 2007). Men are turned out to be more

competitive than women (Samek, 2014). Therefore, men behave differently towards situations where competition plays a major role; men perform better in a competitive environment in comparison to women (Samek, 2014). The timidity of women in heavily competitive positions within organizations is a result from the overconfidence of men (Niederle and Vesterlund, 2007); women feel a low level of confidence in their ability to perform a task and are therefore uncomfortable in a competitive environment (Samek, 2014; Charness et al., 2011). The aversion of risk and feedback plays a less important role (Niederle and Vesterlund, 2007; Samek, 2014). Gender differences towards preferences in competition result in the fact that women have a lower job entry rate, are weak competition to men for promotions and fewer negotiations about salary (Samek, 2014). To conclude, the deviation between self-estimation and real performance is higher for men (Barber and Odean, 2001), men are more competitive (Niederle and Vesterlund, 2007) and men are turned out to handle feedback better (Niederle and Vesterlund, 2007; Barber and Odean, 2001)

Age and life-time experience are the other factors that influence people’s level of self-confidence (Orth et al., 2010), where people’s self-esteem and self-awareness increase with age and experience.

2.3 Existing studies

Several psychological studies show the influence of age and gender in relation to people’s overestimation of actual performance. For example, these studies test overestimation with questionnaires (Eccles et al., 1993; Mathias et al., 2011), estimation games (Newman, 1984) and memory games (Finn and Metcalfe, 2014; Lipko et al., 2012). These results lead to my

overall conclusion that it is important to take gender and age into account in experimental design.

However, the effect of social visibility on stated confidence is not examined in economic studies, with the exception of the paper of Ewers and Zimmermann (2015). My view and study are somewhat different, especially in the following points. Firstly, the paper of Ewers and Zimmermann (2015) focuses on the effect of social image on people’s stated confidence level by testing on relative confidence in a between-subject experimental design. Relative confidence could be explained by definition two of Moore and Healy (2008) mentioned in Section 2.2.1: estimating your individual ability as “above average”. Although some elements of the study of Ewers and Zimmermann (2015) are similar to the view in this thesis (such as that social image could be an explanation of the overestimation of people’s stated confidence in case of reporting private information about personal abilities) this thesis has a slightly different view on the causal relation between overestimation, social image and overconfidence, as well as a difference in the experimental design. The just mentioned causal relation between overestimation, social image and overconfidence in this thesis is explained in table 1 of Section 2.2.2. In short, individuals may overestimate their personal abilities or skills due to social image or overconfidence; this is in contrast with the view of Ewers and Zimmermann (2015) that overconfidence is the cause of protecting people’s social image. Besides, the experimental design of this thesis includes a within-subjects design instead of the between-subject design of Ewers and Zimmermann (2015). Whether the stated confidence of individuals becomes social visible or not is an essential aspect in the reaction of people’s behaviour (Levitt and List, 2007). Therefore I prefer to use a within-subject design to investigate the real effect on people’s stated confidence by increasing social visibility.

Secondly, in this study I try to investigate absolute confidence instead of relative confidence which is used in the study of Ewers and Zimmermann (2015). The effect of absolute stated confidence can be explained by definition one of Moore and Healy (2008) mentioned in Section 2.2.1: absolute estimating of personal skills and abilities. The test is based on cardinal and not ordinal scales. By using absolute confidence, the subjects write down their individual stated confidence scores based on estimation of their own actual performance. The subjects of the study of Ewers and Zimmermann (2015) only had two options to choose their self-reporting from (“above average” or “under average”), where the average of actual performance of the group is unknown to the subjects. In this case the subjects can experience some difficulties with the omitted option of choosing the

troubles with their own personal self-estimation; they will have difficulties with the

estimation of the expected performance of others (Moore and Healy, 2008). Using different measures of stated confidence (relative versus absolute) can lead to contrary results (Moore and Healy, 2008).

Finally, the study of Ewers and Zimmermann (2015) has an anonymous setting between subjects. Experience of social image and – pressure is stronger if individuals are (somehow) connected to each other (Levitt and List, 2007). The experiment in this study is very different, because of the class-setting: the subjects that know each other for at least 5 years.

This Literature Review discussed more relevant studies on the concepts of stated confidence and social visibility besides the important study of Ewers and Zimmermann (2015). The papers discussed on social image mainly focused on people’s image concerns in the specific case of charitable giving (Bateson et al., 2006, Bénabou and Tirole, 2006; Jones and Linardi, 2012). However, people’s motivations to contribute to charities in case of increasing publicity, explained by Bénabou and Tirole (2006), can be extended to people’s motivations to the personal level of stated confidence; people’s concern about image increases in case of social visibility if people attach high value to their image and reputation and this influences individual’s decision-making. Those individuals feel shame by actions or decisions that are below average behaviour of others (Bénabou and Tirole, 2006). The theory of

Bénabou and Tirole (2006) is not restricted to the specific setting of charities and public goods. The theory, of the same authors, why individuals go for a positive self-image (Bénabou and Tirole, 2002) can in my view also be applied to stated confidence.

But this does not hold for every study. For example, the results of the experiments on social image of Bateson et al. (2006) and Jones and Linardi (2012) are difficult to generalize to other settings than charities. The results of social visibility vary between different studies. The effect of monitoring behaviour can be seen as an alternative of increasing (social)

visibility of people’s behaviour. The adjustment in behaviour due to monitoring is negative in case of the study of Falk and Kosfeld (2006), but positive in case of the study of Bateson et al. (2006).

The results of a principal-agent setting on performance of Falk and Kosfeld (2006) is more relevant for this thesis, where performance of individuals also play a major role. However, the person who observes people’s behaviour is an important factor in the decision of the level of stated confidence. In this thesis the observer will be an equivalent person, whereas the observer is a superior in case of the study of Falk and Kosfeld (2006).

Gneezy (2005) is important in my view. Lying-aversion at the other end of social image is the basis for the trade-off between the two, in cases where individuals are aware of their true performance (level). Gneezy (2005) is correct to remark that this lying-aversion holds only unless the negative consequences of lying are zero or neglected.

Barber and Odean (2001) showed a gender difference in case of overconfidence; people are overconfident in general, but men in general have a higher level of overconfidence in comparison to women. Barber and Odean (2001) studied this effect in the stock market, but their results of overconfidence are confirmed by other studies in different settings as well (Niederle and Vesterlund, 2007; Möbius et al., 2011). Overconfidence is often studied in psychology, whereas the study of Barber and Odean (2001) tested overconfidence in the economic realm.

From the sections above I expect that social visibility, gender and age influence

people’s level of stated confidence. I expect that, in case of the trade-off between social image and lying aversion, the value of a high/positive social image in general shall dominate the individual’s desire not to lie. In addition, I expect that individuals are in general overconfident about their personal abilities.

3. Methodology

In the ideal world I prefer to collect my data for studying the effect of social visibility on stated confidence from a firm. But we don’t live in an ideal world, I expected that the cooperation level of such firms to be relatively low for outsiders. Therefore, I developed an experiment at a primary school, where I have a high degree of control on the environment and good cooperation.

The study of Haun and Tomasello (2011), as well as Bronfenbrenner (1970), showed that the sensitivity towards a positive social image starts already at a really young age of the primary school. This makes it possible to investigate the research question with the results in young children at the primary school.

3.1 Experimental design and procedures

3.1.1 Background information of the data

The primary school ‘De Kleine Reus’ in Amsterdam gave me the opportunity and permission to run the controlled classroom memory game experiment in the 5th and 6th grade (Dutch education: group 7 and 8). The experiment was on 12th and 13th of May 2015.The primary

school has two classes of the 5th grade and two classes of the 6th grade. This gave me access to a group of 104 children in the age between 10 – 13 years, divided over four classes in total.

The students of the primary school ‘De Kleine Reus’ do know each other perfectly well within their class. The class composition is the almost the same since the 1rst grade (Dutch education: group 3). This setting can increase the effect of social visibility in this group of subjects in comparison to an anonymous setting where individuals do not know each other; the opinion of a fellow pupil could count more than an opinion of a complete stranger. Besides, the pupils see each other quite often; this could increase the individual’s desire to build up a good reputation within the group (Levitt and List, 2007). However, I expect that the real reputation – and social image building is already defined during their past years at school.

3.1.2 Procedures and experimental design

The members of the ethical commission gave me the permission to run this experiment on 22th of May 2015. The approval letter is included in Appendix 1.

Procedures

All documents of the experiment are written in Dutch. The experiment in my study consists out of the following small parts:

1. An introductory part

2. Two performance measures based on two memory games 3. Two individual performance estimations

4. A small questionnaire.

This experiment is a within subject design with two different types of treatments. The difference between the treatments will be explained in Part three: two individual performance

estimations.

All subjects start with an introductory part. An example of the introduction could be found in Appendix 2. I read the instruction out loud. The subjects were motivated by the announcement of winning a reward at the end of the experiment. The subjects did not know how – and what they could win, only the fact that they should focus and try the best they can on their tasks. The incentive was a small funny eraser. The introduction includes a small and simple memory game to illustrate the essentials of the experiment to follow.

Part two of the experiment included two performance measures based on two memory

games. These performance measures will be called ‘tasks’ in the remaining part of this thesis. One task included twelve random two-digit numbers, while the other task included two-paired alphabet letters. The two tasks are illustrated in table 2 and 3. Whether a pupil first starts with the number or alphabet letter task depends on which class (s)he was in. The procedures per specific class are explained later in this section.

Table 2: Two-digit number task.

65 41 48

46 88 21

71 28 92

79 32 53

Table 3: Two-paired alphabet letter task.

NW MA LQ

PR AM DT

QM TC SO

CV VQ JU

The subjects had 30 seconds to focus on the twelve random numbers or alphabet letters. After that, the subjects had to write down the numbers or alphabet letters they still remembered within 20 seconds.

I tested the stated confidence of the pupil involved with individual performance estimations tasks in Part three of this experiment. I asked the subjects to write down how many answers they did respond right immediately after the two performance tasks. As

mentioned before, this study is a within subject design with two different types of treatments. Based on the information revealed before the subjects have to write down their stated

confidence. The information revealed before writing down the personal stated confidence level depends on which treatment he/she is in. In one treatment I asked the subjects to show one stated confidence level to their classmate sitting next to them. This increases the social visibility on stated confidence and this treatment is called the interpersonal treatment. In the other treatment the stated confidence remained private, without other classmates seeing each other’s performance estimations. This treatment is called the intrapersonal treatment.

Children knew beforehand whether or not their guess will remain private. Whether I find a difference between these two treatments, with or without the effect of social visibility, could be evaluated only by me after collecting the forms with responses; the classmates involved

only saw one of the individual level of stated confidence and will never get an insight into the real performance of their peers.

Every class gets exactly the same two tasks. I only intended to switch in the sequence of the two tasks and in which treatments the subjects started. The details about the sequence of tasks and treatments per specific class are illustrated in table 4. The 5th grade started with the two-digit number task, whereas the 6th grade started with the two-paired alphabet letter task. Besides, one of each grade started with the intrapersonal treatment, whereas the remaining class of the same grade started with the interpersonal treatment.

Table 4: Sequence of tasks and treatments per specific class. The numbers between the brackets are the classroom definitions following the Dutch education program.

5th grade (7a) 5th grade (7b) 6th grade (8a) 6th grade (8b)

Intrapersonal Numbers Interpersonal Numbers Intrapersonal Letters Interpersonal Letters Interpersonal Letters Intrapersonal Letters Interpersonal Numbers Intrapersonal Numbers

The study finished with a very small questionnaire. The questionnaire could be found in Appendix 3. The questionnaire included questions such as ranking the difficulty of the game and performance relative to classmates, as well as information on gender and age.

I asked the subjects to put their forms with answers in an individual blank envelope when the experiment was finished. With this procedure of collecting the data, the responses of the subjects remained completely anonymous, and could not been traced back to the

individual students. Furthermore I have made the promise to the students to evaluate their personal data with severe confidence (also the teachers involved get no insight in the material, only in the combined results on class-level).

Experimental design

The reason why I use numbers and alphabet letters in my experiment is because of a small gender difference with respect to mathematics and Dutch proficiency in the Netherlands (Driessen and Langen, 2010). Boys perform better in mathematics in comparison to girls, while girls are, on the average, better in Dutch proficiency (language and reading) (Driessen and Langen, 2010). It could influence the results if I only had used one of the two types of memory games even if this effect is really small. Although this memory game experiment was not specific about mathematics or language, the use of numbers could be preferred by boys, while girls could prefer the use of alphabet letters. Both tasks should have the same degree of

difficulty, because the design, except for using numbers and alphabet letters, is exactly the same.

This thesis investigates subject’s stated confidence by reporting confidence after the task is finished. In this way I avoid the situation that subjects stop the task when they have reached their reported level of self-estimation. The use of more than one number or alphabet letter on a specific spot increases the difficulty of the task relative to using a more simple memory game with pictures of objects for example. The task should not be too easy to test for stated confidence, because I do not believe that subjects will write down names of objects that were not part of the picture memory game in the first place. I think that this will be different with two-digit numbers and two-paired alphabet letters; children could maybe get confused with the different numbers/letters under “time pressure” of the memory game. It is found in literature, that people could only remember seven things on average (Miller, 1994). The use of twelve random factors makes these memory game tasks really difficult. Due to this the subjects will not know for sure how many answers they will have correct after each part in this memory game experiment with twelve solutions. This is very important in my study: the subjects should make a trade-off between reporting confidence in terms of the number of answers they know for sure on the one hand and reporting confidence to protect their social image in the case of social visibility (interpersonal treatment) on the other hand.

With these procedures and experimental design I tested the stated confidence of individuals.

3.2 Hypotheses

The focus of this experiment is to test whether social visibility influences stated confidence, which is the effect of the two different treatments. The fact that other people will find out about the subject’s possible under-/overestimation of performance estimations could influence their report of self-estimation with respect to self-esteem. Hypotheses are that the

overestimation of subjects is higher in case of the interpersonal, social visible, treatment. Moreover, the effect of gender and age on stated confidence will be measured as well. Driessen and Langen (2010) show gender difference in non-cognitive skills; boys have a higher self-esteem and are more motivated by performance competition relative to girls in the Netherlands. These characteristics of young boys correspond to the characteristics of men explained in the literature review. The non-cognitive skills posture, self-esteem and

performance competition, develop further when boys are getting older (Driessen and Langen, 2010). There is also some evidence that the brain of boys and girls develop differently in the

age of 12-15 years while it is supposed to be equal at the age of primary school groups

(Claessen, 2013). This finding suggests that there should be no gender difference on the actual performance between boys and girls in my experiment.

Also, in my experiment, the actual performance of the tasks should be different based on the difference level of age within the subject group. I assume that older students perform better in comparison to younger students.

I do expect a gender – and age difference in the self-awareness of one’s own responding; older boys in primary schools are better self-aware of their own responding in comparison to girls, but older boys will overestimate their own responding more often than girls. Based on the research question and literature review, the following hypotheses will be tested:

Hypothesis 1: Stated confidence increases in case of introduction of social visibility. Hypothesis 2: Stated confidence is higher for boys in comparison to girls.

Hypothesis 3: Stated confidence is higher for older pupils in comparison to younger pupils.

4. Results

This chapter describes the results of the controlled memory game experiment studied in this thesis. The three hypotheses, mentioned in Chapter 3, will be tested in this Chapter (4). The main topic of the three mentioned hypotheses above is stated confidence. Individual’s stated confidence is related to the actual performance of subjects. For the actual performance I expect the following outcome:

 There is no difference in actual performance based on gender.  The actual performance is higher for older pupils.

Different elements of the experiments could influence subject’s level of stated confidence. These elements are the two different treatments, tasks and sequence of games. Actual performance and stated confidence could differ across game one and game two. The details about this reasoning will be explained in Section 4.2. For the stated confidence I expect the following outcome:

 The stated confidence is higher in case of the interpersonal treatment in comparison to the stated confidence of the intrapersonal treatment (Hypothesis one).

 There is no difference in the level of stated confidence between the two different tasks.

 The stated confidence of game two is higher in comparison to the stated confidence of game one (learning effect).

This should be taken into account for understanding the tables and tests used in this chapter.

4.1 Check for a normal distribution

The following assumptions of Stock and Watson (2012) hold for all factors that will be discussed in this thesis, except for the factor Age:

 Random assignment of the subjects to the different treatments results in independent observations. This entails that the mean of the error term, given the treatments, is equal to zero.

 Large outliers of the observations are removed from the experiment. This creates that large outliers are unlikely in this study.

 The Central Limit Theorem holds; there are more than 30 observations. This also holds for the used subgroups to test for stated confidence. This indicates that the sample means are normally distributed with equal variances between groups.

The factor Age includes subgroups with less than 30 observations. Therefore the Central Limit Theorem is violated in case of Age. For all other factors, these assumptions should confirm that the sample distribution has a normal distribution. To prove that the above assumptions of Stock and Watson (2012) holds in this experiment I test all remaining factors for normality.

Two aspects of a normal distribution are skewness and kurtosis. The results of the skewness indicate the distribution’s symmetry and the results of the kurtosis include

information about the shape of the distribution’s tails (Stock and Watson, 2012). The value of skewness is equal to zero and the value of kurtosis is equal to three in case of a normal

distribution (Stock and Watson, 2012). Table 5 shows the values of Skewness and Kurtosis of all factors of stated confidence used in this thesis. The Joint P-value of the results of the skewness and kurtosis together indicates whether the distribution is normal in table 5. A Joint

P-value is smaller than 0,1 rejects that the sample distribution is normally distributed. The

sample distribution of Girls in the Interpersonal treatment is not normally distributed at a 10% significance level. The negative skewness of Girls in the Interpersonal treatment indicates that the distribution is not symmetric at a 5% significance level. The sample distribution of Girls in the Interpersonal treatment has a longer tail on the left-hand of the distribution in comparison to the right-hand side of the distribution. The results of the Joint

P-value of the remaining factors in table 5 confirm that the factors are normally distributed.

Table 5: Test for normality: Stated Confidence. ** and * indicate statistical significance at 5% and 10% respectively using a two-sided test.

Stated Confidence Treatment Observations Skewness Kurtosis

Joint P-value All subjects Interpersonal 103 -0,188 2,933 0,701 Intrapersonal 103 0,334 2,779 0,336 Gender Boys Interpersonal 49 0 2,222 0,361 Girls 51 -0,634** 3,85 0,052* Boys Intrapersonal 49 0,158 2,722 0,879 Girls 51 0,491 2,979 0,254 Class 6th grade Interpersonal 54 -0,388 2,467 0,329 5th grade 49 -0,227 4,23* 0,13 6th grade Intrapersonal 54 0,407 2,927 0,387 5th grade 49 0,45 2,682 0,355 Task Numbers 103 0,126 2,349* 0,188 Letters 103 -0,041 3,405 0,533 Numbers Interpersonal 52 0,016 2,433 0,72 Letters 51 -0,407 3,453 0,229 Numbers Intrapersonal 51 0,247 2,256 0,287 Letters 52 0,557* 3,065 0,175 Game One 103 0,335 2,718 0,317 Two 103 -0,19 2,984 0,675 One Interpersonal 52 0,456 2,785 0,294 Two 51 -0,545* 2,919 0,212 One Intrapersonal 51 0,39 2,666 0,44 Two 52 0,467 2,597 0,334

Table 6 shows the results of the test for normality in case of the Actual Performance of the subjects. The Joint P-value of the Interpersonal and Intrapersonal treatment indicates that the treatments are not normally distributed at a 10% significance level. The distribution of the

Interpersonal treatment is less heavy tailed in comparison to a normal distribution at a 10%

significance level, whereas the distribution of the Intrapersonal treatment is not symmetric with a longer tail on the right-hand side of the distribution in comparison to the left-hand side of the distribution.

Table 6: Test for normality: Actual Performance. * indicate statistical significance at 10% using a two-sided test.

Actual performance Observations Skewness Kurtosis Joint P-value

Treatment Interpersonal 103 -0,33 2,34* 0,08* Intrapersonal 103 0,435* 3,475 0,088* Task Numbers 103 0,178 2,632 0,589 Letters 103 -0,257 2,644 0,434 Game One 103 0,1 2,565 0,612 Two 103 0,028 2,862 0,992

A Paired T-test will be used to check whether the average stated confidence of subjects differs between the two different treatments or whether the subjects overestimate their actual

abilities. An Unpaired (two sample) T-test will test whether there are differences in

performance for different groups of subjects. The T-test can only be used in case of a normal sample distribution. The Wilcoxon Signed Rank Sum Test is a substitution for the Paired T-test and the Wilcoxon Rank Sum Test is a substitution for the Unpaired (two sample) T-T-test in case of the nonnormal distribution. The Wilcoxon Signed Rank Sum Test and Wilcoxon Rank Sum Test are nonparametic tests of the difference in locations instead of difference in average (Keller, 2009). The nonparametic test will be used for the factors Age and Girls in the

Interpersonal treatment.

4.2 Descriptive data

This section explains the results of the responses of the questionnaire and the actual

removed the observations of one subject due to the unreliable answers of his stated

confidence, such as 100 and 1000 answers correct. Including this outlier would lead to biased results.

The result is a study with a remaining number of 103 subjects. The characteristics of the subjects are illustrated in table 7. The group consists of 52% girls and 48% boys; which is approximately an equal distribution. This also holds for the distribution of subjects over the different classes; 48% of the subjects are in the 5th grade and 52% of the subjects are in the 6th grade. The subjects have an age between 10 to 13 years and with an average age of 11 years. The total group of 103 subjects consists of four subgroups with earlier described different sequence of the two tasks (numbers versus alphabet letters) and the two treatments (intra- versus interpersonal) in table 4. The summary characteristics per class could be found in Appendix 4.

Table 7: Characteristics of subjects. The numbers between the brackets are standard errors. The t-test *** indicate statistical significance at 1% by using a(n):

a _{two-sided test} b _{one-sided test}

The column Total Performance in table 7 includes the average of correct answers over the two tasks together by All Subjects, Gender, Age and Class. The column P-value includes the outcome of the T-test. This column only includes a t-test for Gender and Class, where

Class is used as a factor for age. The Total Performance of All Subjects is 7,039. Girls

perform slightly better in comparison to Boys. However, the difference in Total Performance, based on Gender, is not significant (0,755). This confirms the literature of Cleassen (2013);

Observations Percentages Mean Std

Total Performance P-value All Subjects 103 7,039 Gender Boys 49 48% 6,959 (0,401) 0,755a Girls 54 52% 7,111 (0,283) 103 0,5243 0,5019 Age 10 25 24% 6,64 11 49 48% 7,184 12 26 25% 7 13 3 3% 8,333 103 11,068 0,7829 Class 5th 49 48% 6,367(0,341) 0,004***b 6th 54 52% 5,524 0,502 7,648(0,321) 103

the total performance of children in primary school is the same based on gender. The Total

Performance of subjects in the 6th class is significantly higher than the Total Performance of subjects in the 5th class at 1% significance level; the 6th grade perform significantly better on these two tasks in comparison to the 5th grade.

Figure 1 shows the remaining results of the questionnaire filled out by the subjects after the experiment finished. The remaining results of the questionnaire describes the subject’s opinion about the Degree of Difficulty, their level of Relative Confidence and their opinion about the Most Difficult Task.

Starting with the distribution of Degree of Difficulty (Figure 1A). The distribution of the Degree of Difficulty is approximately the same for Numbers and Letters. No subject found both tasks Really Easy and less than ten subjects found the tasks Easy. Approximately 50% of the subjects found the two tasks Difficult to Really Difficult. The largest group of subjects filled out a Medium level of difficulty. Overall, we can conclude that the subjects reported a high Degree of Difficulty on both tasks.

Figure 1B describes the subject’s responses about their Relative Confidence. The distribution of Relative Confidence of the subjects is also the same for Numbers and Letters. Almost 70 subjects rated their performance “on average” (Medium level) and only 8 subjects rated their performance as “One of the Best in Class”. Based on this information we see a restricted degree of relative overconfidence in case of definition two of Moore and Healy (2008): estimating individual ability as “above average”. None of the subjects thought that they were the “Best in Class”. A total of 26 subjects showed underconfidence in relation to their performance. Underconfidence is indicated as Very Bad and Not Really Good in Figure 1B.

I mentioned in Chapter 3 that both tasks were exactly the same, except for using numbers and alphabet letters. However, the responses of the question about the Most Difficult

Task did not confirm this expectation. The responses of the Most Difficult Task can be found

in Figure 1C. In Figure 1C we can see that only 13 subjects thought that both tasks had the same degree of difficulty. The subjects found the alphabet Letter task slightly more difficult in comparison to the Number task.

Figure 1: Results questionnaire: Degree of Difficulty, Relative Confidence and Most Difficult Task. The total subjects for numbers are 102 and for letters are 103. One subject forgot to fill in the Degree of Difficulty and Relative Confidence for the number part.

A B C 0 10 20 30 40 50 60 Really difficult

Difficult Medium Easy Really easy

Fr e q u e n cy Degree of Difficulty

Degree of Difficulty

Very bad Not really good Medium level One of the best in class Best in class Fr e q u e n cy Relative Confidence

Relative Confidence

Both tasks are equal to each other Numbers Letters Fr e q u e n cy

Most difficult task

4.3 Interpersonal versus intrapersonal

The intra- versus interpersonal treatments illustrate the two different positions of subjects in case of reporting their level of confidence. Subjects report their stated confidence in private in case of the intrapersonal treatment, while subjects should show their report of stated

confidence to a classmate in the interpersonal treatment. The within-subject design indicates that all subjects were part of both treatments; where the hypotheses expect that the stated confidence is higher in case of the interpersonal treatment, for boys and older students. The T-test of the intrapersonal and interpersonal stated confidence is T-tested for the following

categorical groups: All Subjects, Gender, Age, Tasks and Games. The categorical group

Gender is also tested with the nonparametic tests, because the sample distribution of Girls in

the Interpersonal treatment did not have a normal distribution (Section 4.1).

Results of the t-test for the three hypotheses could be found in table 8. The column

Observations summarizes the number of subjects. The column Intrapersonal includes the

average stated confidence in the intrapersonal treatment. The column Interpersonal includes the average stated confidence in the interpersonal treatment. The column P-value includes the outcome of T-test by comparing the difference in means of the intrapersonal and interpersonal treatment. The P-value mentioned in the rows includes the outcome of T-test by comparing the difference in means of the categorical groups per specific treatment.

Table 8: T-test of intra- and interpersonal stated confidence. The numbers between the brackets are standard errors. The t-test ***, **, * indicate statistical significance at 1%, 5% and 10%, respectively by using a(n):

a _{Two-sided test} b _{One-sided test} c

Paired T-test

d _{Unpaired (two sample) T-test}

Observations Interpersonal Intrapersonal P-valueb,c

All subjects 103 4,175 (0,14) 3,981 (0,141) 0,119 Gender Boys 49 4,286 (0,226) 4,204 (0,21) 0,371 Girls 54 4,074 (0,173) 3,778 (0,188) 0,089* P-valuea,d 0,227 0,066* Age 6th grade 54 4,5 (0,204) 4,204 (0,176) 0,121 5th grade 49 3,816 (0,179) 3,735 (0,222) 0,347 P-valueb,d 0,007*** 0,05** Tasks Numbers 51 vs. 52 4,192 (0,2) 3,98 (0,24) 0,249 Letters 52 vs. 51 4,157 (0,199) 3,981 (0,154) 0,242 P-valuea,d 0,45 0,5 Game Game one 51 vs. 52 4,173 (0,177) 3,804 (0,221) 0,097* Game two 52 vs. 51 4,176 (0,22) 4,154 (0,177) 0,469 P-valuea,d 0,99 0,218

The results in table 8 show that the stated confidence is not significantly higher in case of increasing social visibility (0,119) if we look at the All Subjects category. However, if all subjects are divided into two groups by Gender, table 8 shows that the stated confidence of

Girls is significantly higher in case of increasing social visibility at a 10% significance level

(0,089). The results of the nonparametic tests of Gender could be found in table 9. Girls do not behave significantly different in case of increasing social visibility in the nonparametic test. The results of both test are contradicting. However, the p-value of Girls (0,117) is still close to a 10% significance level in the nonparametic test. By looking at the results of the two tests (Paired T-test and Wilcoxon Signed Rank Test) I tend to the conclusion that girls do react more on the different treatments of reporting their stated confidence in comparison to boys, but the results of the two tests of the effect of the interpersonal treatment on girls is ambiguous in case of confirming hypothesis one. Hypothesis one “stated confidence increases

in case of introduction of social visibility” is only confirmed in case of girls at a 12%

significance level.

Table 9: Wilcoxon signed rank test. H0:Stated Confidence Interpersonal > Stated Confidence Intrapersonal. The p-values indicate no statistical significance by using an one-sided test.

P-value Gender Boys 0,414 Girls 0,117 Age 10 0,193 11 0,151 12 0,368 13 0,391

The stated confidence of boys is higher in comparison to girls in both treatments (Mean: 4,204 > 3,778 and 4.286 > 4,074). However, the stated confidence of Boys is only significantly higher relative to Girls in case of the Intrapersonal treatment at a 10% significance level. The nonparametic test in table 10 confirms this finding of the Unpaired (two sample) T-test. This could be explained by the fact that only girls react differently to treatments after social visibility; the fact that girls should show their self-estimation to others increase their stated confidence score nearly to the level of stated confidence of boys.

Whether the subject first gets the intra- or interpersonal treatment depends on in which class he/she is in. All four classes are included in this case.

Table 10: Wilcoxon rank sum test. H0: Stated Confidence Boys > Stated Confidence Girls. * indicate statistical significance at 10% by using an one-sided test.

P-value

Treatment

Interpersonal 0,311

Intrapersonal 0,057*

It could also be possible that the actual performance by gender could play a role in explaining the difference in stated confidence in case of the intrapersonal treatment. Girls performing significantly better in the interpersonal treatment could explain the increasing level of stated confidence in table 8. However, we saw that the total performance of boys and girls were the same in Section 4.1. There is a remaining possibility, that the actual performance per specific task could be different among gender in case of different sequence of the two treatments defined in table 4. Whether boys actually perform better in comparison to girls per Treatment,

Task and Game will be tested in the following Section 4.3. For now, hypothesis two “Stated confidence is higher for boys” is only confirmed in case of the intrapersonal treatment.

For hypothesis three (stated confidence is higher for older pupils in comparison to

younger pupils) I use the Wilcoxon Rank Sum Test, because the subgroups of Age are smaller

than 30 observations. Table 7 showed that there were only three observations of subjects with an age of 13. Table 11 shows the results of hypothesis three excluding the observations of subjects with an age of 13. The results of table 11 show that the stated confidence is

significantly higher for subjects with an age of 12 in comparison to subjects with an age of 10 at a 5% significance level, as well as in comparison to subjects with an age of 11 at a 10% significance level in the intrapersonal treatment. Table 9 shows that there is no significant difference in the stated confidence of the inter- and intrapersonal treatment of subject groups of the same age.

I can use the results of the T-tests in table 8 if I define the two levels of Class(es) as a factor of experience; the average life-time experience of the 6th grade is higher than the average life-time experience of the 5th grade. I want to test whether these two groups behave significantly different from each other by using the two levels of classes (5th and 6th) where

Class is normally distributed. The results in table 8 show that the stated confidence of the 6th

grade is significantly higher in comparison to the 5th grade in both treatments at a 5% and 1% significance level. There is an additional, significant difference of stated confidence of “older” (or one year more educated) subjects in the interpersonal treatment in case of using

indicator of the factor age. Therefore hypothesis three “Stated confidence is higher for older

pupils in comparison to younger pupils” could be confirmed only for the intrapersonal

treatment.

Also in this case, it could be possible that the actual performance by age could play a role in the level of stated confidence. Whether older subjects actually perform better in

comparison to younger subjects per Treatment, Task and Game will be tested in the following

Section 4.4.

Table 11: Wilcoxon rank sum test, Age. ** and * indicate statistical significance at 5% and 10% respectively by using an one-sided test.

Interpersonal Intrapersonal

Age 10 11 10 11

10

11 0,273 0,302

12 0,221 0,406 0,034** 0,081*

I also included the two different types of tasks in table 8, because the subjects reported in the questionnaire that the degree of difficulty of both tasks was not the same (Figure 1C). However, there is no significant effect found on the stated confidence based on these two tasks.

Furthermore, I included the factor Game into table 8 as well. The subjects played the game twice. Therefore it is possible that the subjects experience a learning effect and

performed better the second time. The subjects incorporate this effect and increase their stated confidence in game two. This leads to the possibility that the stated confidence of game one is smaller in comparison to the stated confidence of game two. From the results in table 8 we see that the subjects did not significantly perform better in Game Two (0,109 and 0,495). Even though, the difference of stated confidence in case of the Intrapersonal treatment is very close to a 10% significance level. However, I have found a significant difference between the two treatments in Game One. There were 52 subjects that were part of the Intrapersonal – Interpersonal treatment and 51 subjects that were part of the Interpersonal – Intrapersonal treatment. I found a general conclusion that the stated confidence of the group of subjects that started with the Interpersonal treatment is significantly higher than the stated confidence of subjects that started with the Intrapersonal treatment at a 10% significance level. Appendix 5 takes a closer look to the difference in stated confidence of the two different treatments within the following specified groups of subjects; intrapersonal – interpersonal and interpersonal –

intrapersonal. There is no significant difference by comparing the stated confidence levels within the group of subjects that started with the Intrapersonal treatment and finished with the

Interpersonal treatment (mean: 3,804 and 4,176), but the p-value of 0,118 is close to a 10%

significance level (Appendix 5). Starting with the Interpersonal treatment and finishing with the Intrapersonal treatment is not significant as well. This could be an important result. Starting with the social visible treatment of reporting stated confidence leads to higher levels of stated confidence in the private treatment. Higher levels of stated confidence tend to be transported to the next game. This could be relevant for group processes in real life.

The conclusion of the last two paragraphs together illustrates the following results: the effect of treatments on stated confidence is significant in the first game and the difference of stated confidence is higher in case of starting with the intrapersonal treatment.

The conclusion of this section is that hypothesis one is confirmed in case of girls at a 12% significance level, hypothesis two and three is confirmed in case of the intrapersonal treatment.

4.4 Overestimation of subject’s performance

Literature showed that individuals overestimate their abilities in comparison to their actual performances. Appendix 6 shows that less than 10% of the subjects underestimate their performance in my experiment. The remaining 90% is approximately equally divided over estimations and overestimations. Appendix 6 indicates that overestimation plays also a major role in my experiment in contrast to the results of the questionnaire of Relative Confidence (Figure 1B). Table 12 confirms that subjects generally overestimate their abilities, where the self-estimation is significantly higher than the actual performance at a 1% significance level in all cases described table in 12. The overestimation by subjects holds even when the whole group is divided by gender or age (Appendix 7).